An Interaction Pattern Kernel Approach for Protein-Protein Interaction Extraction from Biomedical Literature - Technologies and Applications of Artificial Intelligence - page 43

where #ch(n 1 ) is the number of children of node n 1 ; ch(n, k) is the k th child of node n ;
and λ (0< λ <1) is the decay factor used to make the kernel value less variable with
respect to different sized sub-trees. The parse tree kernel counts the number of
common sub-trees as the syntactic similarity measure between two relation instances.
The time complexity for computing this kernel is

O ( N ⇅ N )

.

1 2

5 Experiments

5.1 Experimental Setting

We evaluated our method with three publicly available corpora that contain PPI
annotations: LLL [13], IEPA [4] and HPRD50 [6] (the distribution of corpora are
shown as the Fig.5). All the corpora are parsed using Stanford parser
(http://nlp.stanford.edu/software/lex-parser.shtml) to generate the output of parse tree
and part-of-speech tagging. In our implementation, we use Moschitti's tree kernel
toolkit [1] to develop the convolution kernel of an IPT. To derive credible evaluation
results, we utilize the 10-fold cross validation method [1] on all of the corpora. This
guarantees the maximal use of the available data and allows comparison to the earlier
relevant work. The evaluation metrics are the precision rate, recall rate, and F1-
measure [1]. The F1 value is used to determine relative effectiveness of the compared
methods. We exploit the macro-averaged score to indicate the overall performance
across three different corpora for each evaluation metric.

Fig. 5. Distribution of 3 corpora used for performance evaluation of PPI extraction

5.2 Results and Discussion

The proposed interaction pattern tree kernel uses the PPI patterns to enhance the SPT.
In the following, we compare it with several feature-based and kernel-based PPI
extraction methods reported by [17] to demonstrate the effectiveness. As shown in
Table 1, the proposed method significantly outperforms SPT and AkanePPI.
Furthermore, the syntax tree-based kernel methods (ST, SST, PT, and SpT) only
examine the syntactic structures of text and cannot sense the semantics of protein